NLP Chatbot for Accelerating Internal Operations

Business Overview

The client is a multinational company serving the combined health information technologies and clinical research industries.

The employees interact with the flagship healthcare product daily, and when updates are released for the client’s internal system—documentation is compiled for employees to review. This was a time-consuming process that also affected overall performance.

The client set out to develop a chatbot to help employees quickly find the information they need, making it easier and faster to get tips on the system, saving time on higher-value activities.

Project Scope

Realize a question-answering (Q&A) ML algorithm to answer user’s questions based on the content

Develop a chatbot based on the algorithm available for mobile, web, and desktop

Integrate the chatbot in the client’s system smoothly

Solution

Azure Bot
Text Classification Model
Robust Data Processing

Azure Bot

QA bot is implemented via Bot Framework Python SDK and based on the Azure Bot—a powerful conversational AI from Microsoft. When the bot gets a question from the user, it sends it to the QA service stored within Azure Container Instances and gives a response back to the user.

Users also could provide feedback to the bot via buttons. This feedback is essential for making new training data that will improve the quality of the QA model response. We developed two deployment approaches—a manual one for more flexible configuration and one through the Azure DevOps Pipeline.

The system is divided into two parts:

Server—for training and making predictions
Client-facing UI—user sees only information which the algorithm defines as a minimum of 80% reliable; anything less is irrelevant

Text Classification Model

For this task, the NIX team has chosen the Tensorflow Hub implementation of the widely used BERT natural language model and implemented a classifier. It was a best-fitting solution- robust framework, strong pre-trained language models, and ability to easily apply and modify the Keras interface for Tensorflow.

It also provides the ability to join the pre-processing model with the primary model (so that we can directly send text requests to the Tensorflow Serving Server without any intermediate steps) and allows for an easy way to deploy trained models.

Robust Data Processing

Our practice has shown that a pre-trained BERT is excellent at finding the correct answers to the questions contained in the documents. The difficulty lies directly in processing the documents themselves. Documents are HTML pages with a complex nested structure, which makes it challenging to find the correct answer to a question.

For this purpose, we developed a separate module that parses the documents and converts them into a flat format while preserving their contextual structure and richness. We also used the BERT model to search for the most relevant documents.

Outcome

The client received an Azure-based chatbot empowered with conversational AI for web, mobile, and desktop.

The chatbot simplifies the process of finding relevant information for the employees and saves a lot of effort, resulting in a performance boost and more time for higher-value activities.

Team:

Project Manager, Business Analyst, 3 Data Scientists

Tech Stack:

TensorFlow, PyTorch, Scikit-learn, Spark MLlib, Pandas, BERT QA

Request a consultation

Relevant Success Stories

AWS-powered Development Platform for Clinical Trials Management

NIX built a pioneering AWS solution that blends cost-efficiency, adaptability, and cutting-edge technology.

Navigating the Cloud: Modernization of Healthcare Data Pipelines

By embracing a cloud-centric approach and optimizing data pipelines, NIX empowered healthcare and insurance providers with cost-effective, secure, and data-driven solutions.